prompt-engineering

Efficientworkflow

Mission

Structure → test → refine → version. Every prompt is a versioned, tested artifact.

When to Use

Use when writing a new system prompt, building a prompt template, improving an existing prompt that is failing or hallucinating, versioning prompts, chaining prompts into pipelines, calibrating agent autonomy levels, or evaluating prompt quality.

Triggers: “write a system prompt”, “improve this prompt”, “prompt is hallucinating”, “prompt template”, “chain these prompts”, “calibrate autonomy”, “prompt version”

Skills Invoked

• prompt-engineering — design from scratch
• prompt-chaining — multi-step prompt chains
• prompt-refinement — iterative improvement from eval feedback
• prompt-hierarchy — system → instruction → user layer design
• eval-prompt — score generated prompts

Chain-To

• quality-evaluate — benchmark the new prompt against baseline
• policy-govern — enforce governance layers in the prompt hierarchy

Refinement Loop

prompt → eval-prompt score
   ↓ (below threshold)
prompt-refinement
   ↓
modified prompt → re-score
   ↓ (repeat)

Example

{
  "request": "The code generation prompt keeps producing code with `any` types. Fix it."
}

Output: Diagnosis of failure mode, refined system message with explicit TypeScript constraints, version tag, and eval score comparison.